The present invention relates to a nonvolatile magnetic memory device, and a photomask for use in the manufacture of the nonvolatile magnetic memory device.
Attendant on the drastic spread of personal small apparatuses such as communication apparatuses, particularly, the personal digital assistants, various semiconductor devices such as memories and logical circuits constituting the apparatuses are desired to have higher performance, such as higher degree of integration, higher operating speed, lower power consumption, etc. Particularly, nonvolatile memories are considered to be keenly desired in the ubiquitous computing age. Even in the cases of consumption or trouble in the power supply or in the cases of cutoff between a server and a network due to some disorder, the nonvolatile memory makes it possible to preserve and protect important information. In addition, while the recent portable apparatuses are designed to suppress power consumption as much as possible by putting unnecessary circuit blocks into the stand-by state, if a nonvolatile memory capable of functioning as both a high-speed work memory and a large-capacity storage memory can be realized, it is possible to eliminate the wastefulness in power consumption and memory. Besides, the “instant-ON” function enabling an instantaneous start upon making the power supply can also become possible if a high-speed large-capacity nonvolatile memory can be realized.
Examples of the nonvolatile memory include flash memories using semiconductor materials, and ferroelectric nonvolatile semiconductor memories (FERAMs, Ferroelectric Random Access Memories) using ferroelectric materials. However, the flash memories have the defects that the write speed is on the order of microseconds, which is lower than desired. On the other hand, the FERAM has a number of times of overwriting possible on the order of 1012 to 1014, which is too low for the FERAM to be used in place of SRAM or DRAM, and there is the problem that micro-processing of the ferroelectric material layer is difficult to carry out.
As a nonvolatile memory free of the above-mentioned defects, the nonvolatile magnetic memory devices called MRAM (Magnetic Random Access Memory) has come to be paid attention to. The MRAM in the early stage has been based on a spin valve using the GMR (Giant Magnetoresistance) effect. However, the early MRAM has the drawback that, since the memory cell resistance of the load is as low as 10 to 100 Ω, the power consumption per bit at the time of reading is large, and it is difficult to increase the capacity.
On the other hand, the MRAM using the TMR (Tunnel Magnetoresistance) effect has come to be paid attention to in recent years, since the resistance variation ratio at room temperature has been improved to about 20%, from the values of about 1 to 2% in the beginning stage of development. The TMR type MRAM is simple in structure, promises easy scaling, and has a large number of times of overwriting possible because of the recording by rotation of the magnetic moment. Furthermore, with the TMR type MRAM, a very short access time is expected, and it is said that the TMR type MRAM has already come to be able to operate at a rate of 100 MHz.
A schematic, partly sectional view of the TMR type MRAM (hereinafter referred to simply as MRAM) is shown in FIG. 6. The MRAM includes a tunnel magnetoresistance device TMJ connected to a selection transistor TR including a MOSFET.
The tunnel magnetoresistance device TMJ has a laminate structure of a first ferromagnetic material layer 31, a tunnel insulation film 34, and a second ferromagnetic material layer 35. More specifically, the first ferromagnetic material layer 31 has a two-layer structure of, for example, an antiferromagnetic material layer 32 and a ferromagnetic material layer (also called the anchor layer or magnetization fixation layer) 33, in this order from the lower side, and has a strong unidirectional magnetic anisotropy due to the exchange interaction between the two layers. The second ferromagnetic material layer 35 whose magnetization direction can be rotated comparatively easily is also called a free layer or a recording layer. Incidentally, the second ferromagnetic material layer may be called the recording layer 35 in the following description. The tunnel insulation film 34 plays the roles of interrupting the magnetic coupling between the recording layer 35 and the magnetization fixation layer 33 and of passing a tunnel current. A bit line BL for connection between the MRAMs is formed on an upper interlayer insulation layer 26. A top coat film 36 provided between the bit line BL and the recording layer 35 functions to prevent mutual diffusion between the atoms constituting the bit line BL and the atoms constituting the recording layer 35, to reduce the contact resistance, and to prevent the oxidation of the recording layer 35. In the figure, the reference numeral 37 denotes an extraction electrode connected to the lower surface of the antiferromagnetic material layer 32.
Further, a write word line RWL is disposed on the lower side of the tunnel magnetoresistance device TMJ, with a lower interlayer insulation layer 24 therebetween. Incidentally, the extension direction of the write word line RWL (first direction) and the extension direction of the bit line BL (second direction) are ordinarily orthogonal to each other.
On the other hand, the selection transistor TR is formed at a portion of a silicon semiconductor substrate 10 surrounded by a device isolation region 11, and is covered with an interlayer insulation layer 21. A source/drain region 14B on one side is connected to an extraction electrode 37 for the tunnel magnetoresistance device TMJ through a contact hole 22 including a tungsten plug, a landing pad portion 23, and a contact hole 25 including a tungsten plug. A source/drain region 14A on the other side is connected to a sense line 16 through a tungsten plug 15. In the figure, reference numeral 12 denotes a gate electrode, and reference numeral 13 denotes a gate insulation film.
In the MRAM array, the MRAM is arranged at each of the intersections (overlapping regions) in the lattice composed of the bit lines BL and the write word lines RWL.
In writing data into the MRAM configured as above, a current in the positive or negative direction is passed through the bit line BL, while a current in a fixed direction is passed through the write word line RWL, and the composite magnetic field thus generated changes the magnetization direction of the second ferromagnetic material layer (recording layer 35), whereby “1” or “0” is recorded in the second ferromagnetic material layer (recording layer 35).
On the other hand, reading of data is conducted by setting the selection transistor TR into the ON state, passing a current through the bit line BL, and detecting via the sense line 16 the change in the tunnel current due to the magnetoresistance effect. Where the magnetization directions of the recording layer 35 and the magnetization fixation layer 33 are equal, a low resistance result is obtained (this state is made to be “0”, for example), and where the magnetization directions of the recording layer 35 and the magnetization fixation layer 33 are anti-parallel, a high resistance result is obtained (this state is made to be “1”, for example).
In reading the data, suppression of the dispersion in the resistance of the recording layer 35 by maximizing the uniformity of the areas (projection areas) of the tunnel magnetoresistance devices TMJs leads to a reduction in the data reading dispersion, whereby the yield of manufacture is enhanced. One example of the distribution of resistance of the recording layer 35 is shown in FIG. 19. By making the resistance variation ratio as uniform as possible and suppressing the variance width (dispersion) of resistance, it is possible to obtain a larger operation margin of the MRAM and to achieve a higher yield of manufacture. In other words, for the same operation margin on a design basis, it is possible to obtain a larger signal voltage and a higher-speed operation.
On the other hand, in writing the data, suppression of the variance width (dispersion) of the switching magnetic fields (HSwitch) of the tunnel magnetoresistance devices TMJs may be indispensable for obtaining a large-capacity memory.
FIG. 20 shows an asteroid curve of MRAM disclosed in U.S. Pat. No. 6,081,445. Currents are passed through the bit line BL and the write word line RWL, and, based on the composite magnetic field thus generated, data is written into the tunnel magnetoresistance device TMJ constituting the MRAM. The write current flowing through the bit line BL forms a magnetic field (HEA) in the easy axis of magnetization of the recording layer 35, and the current flowing through the write word line RWL forms a magnetic field (HHA) in the hard axis of magnetization of the recording layer 35. Depending on the configuration of the MRAM, there may be the cases where the write current flowing through the bit line BL forms the magnetic field (HHA) in the hard axis of the recording layer 35, and the current flowing through the write word line RWL forms the magnetic field (HEA) in the easy axis of the recording layer 35.
The asteroid curve shows a threshold for inversion of magnetization direction of the recording layer 35 due to the composite magnetic field (composition of magnetic field vectors of the magnetic field HHA and the magnetic field HEA exerted on the recording layer 35), and, when a composite magnetic field corresponding to the outside (OUT1, OUT2) of the asteroid curve is generated, inversion of the magnetization direction of the recording layer 35 occurs, whereby data is written. On the other hand, when a composite magnetic field corresponding to the inside (IN) of the asteroid curve is generated, inversion of the magnetization direction of the recording layer 35 does not occur. In addition, to the MRAMs other than that at the intersection between the write word line RWL and the bit line BL through which the current flow, a magnetic field generated by the write word line RWL or bit line BL alone is applied, so that if the magnitude of this magnetic field is not less than the switching magnetic field (HSwitch) [the region (OUT2) on the outside of the broken lines in FIG. 20], the magnetization direction of the recording layer 35 constituting the other MRAMs than that at the intersection would also be inverted. Therefore, only in the case where the composite magnetic field is on the outside of the asteroid curve and in the region (OUT1) on the inside of the broken lines in FIG. 20, selective writing into the selected MRAM can be achieved.
More specifically, in writing the data, as has been described above, currents are passed through the bit line BL and the write word line RWL to generate a composite magnetic field; in this case, the magnitude of the composite magnetic field is set to be located slightly on the outside of the asteroid curve, and about one half of the composite magnetic field is generated by use of the bit line BL and the write word line RWL. Incidentally, such a state as this is called a “half selected” state. By achieving such a “half selected” state, data is written into the tunnel magnetoresistance device TMJ located at the intersection between the bit line BL through which the current flows and the write word line RWL through which the current flows. On the other hand, no data is written into the tunnel magnetoresistance devices TMJs located at the intersections between the bit lines BLs through which the current flows and the write word lines RWLs through which no current flows, or the tunnel magnetoresistance devices TMJs located at the intersections between the bit lines BLs through which no current flows and the write word lines RWLs through which current flows. Incidentally, such a tunnel magnetoresistance device TMJ (into which no data is written) will be referred to as a non-selected tunnel magnetoresistance device TMJ, for convenience. (see U.S. Pat. No. 6,081,445, U.S. Patent Publications Nos. 6,545,906 B1 and 6,633,498 B1, and S. S. Parkin et al, Physical Review Letters, 7 may, pp.2304-2307 (1990))